Search for: All records

Treadmill running is a common workout for individuals across fitness levels. In this paper, we propose mm-RunAssist, a first-of-its-kind mmWave-based system that enhances treadmill workouts by monitoring respiration waveforms, running rhythm (i.e., coordination between breathing and strides), and detecting fall-off events. Extracting respiration from moving subjects using RF signals is challenging due to dominant motion artifacts. While prior deep learning efforts use adversarial or contrastive learning to mitigate such artifacts, they have been evaluated primarily under low-intensity activities like walking. To address this gap, mm-RunAssist introduces a Dual-task Variational U-Net that shares latent representations between respiration and upper-body movement tracking. This dual-task setup, guided by belt and depth sensors during training, improves reconstruction under intense body motion. Our system not only recovers fine-grained respiratory patterns during running but also supports cadence analysis through arm swing tracking. Extensive experiments with three state-of-the-art baselines under various conditions demonstrate mm-RunAssist's robustness and accuracy in treadmill running scenarios. Results show that mm-RunAssist advances RF sensing by effectively extracting vital signs even during vigorous body movements, offering new capabilities for fitness monitoring and non-intrusive health assessment. 

Accurate and real-time gait analysis is essential for enhancing performance and reducing injury risks in treadmill running. In this paper, we introduce VibRun, an unobtrusive gait analysis system that estimates key physiological metrics, such as cadence, ground contact time, stride time, center of pressure, and plantar pressure distribution, through footstep vibrations captured by low-cost treadmill-mounted sensors. Leveraging advanced multi-task transformer models, our system offers a robust, real-time solution to monitor and analyze running biomechanics without requiring intrusive wearable devices. This approach enables seamless integration into virtual sports, gaming platforms, and immersive exercise environments, enhancing the running experience by providing personalized feedback. By offering precise biomechanical insights in real-time, VibRun paves the way for future applications in virtual sports, gamified fitness, and interactive training programs, empowering users to engage more effectively in their training sessions while improving overall performance and reducing injury risks. Extensive evaluations with 17 participants across varied treadmill running scenarios demonstrate VibRun's accuracy in real-time gait analysis. For instance, VibRun achieves a mean error of 28.8 ms in ground contact time and a distance of 13.66 mm in the center of pressure, among other measured metrics, highlighting its precise performance across multiple gait parameters. 

Federated Learning (FL) is a technique that allows multiple parties to train a shared model collaboratively without disclosing their private data. It has become increasingly popular due to its distinct privacy advantages. However, FL models can suffer from biases against certain demographic groups (e.g., racial and gender groups) due to the heterogeneity of data and party selection. Researchers have proposed various strategies for characterizing the group fairness of FL algorithms to address this issue. However, the effectiveness of these strategies in the face of deliberate adversarial attacks has not been fully explored. Although existing studies have revealed various threats (e.g., model poisoning attacks) against FL systems caused by malicious participants, their primary aim is to decrease model accuracy, while the potential of leveraging poisonous model updates to exacerbate model unfairness remains unexplored. In this paper, we propose a new type of model poisoning attack, EAB-FL, with a focus on exacerbating group unfairness while maintaining a good level of model utility. Extensive experiments on three datasets demonstrate the effectiveness and efficiency of our attack, even with state-of-the-art fairness optimization algorithms and secure aggregation rules employed. We hope this work will help the community fully understand the attack surfaces of current FL systems and facilitate corresponding mitigation to improve their resilience. 

This paper summarizes the main results and contributions of the MagNet Challenge 2023, an open-source research initiative for data-driven modeling of power magnetic materials. The MagNet Challenge has (1) advanced the stateof-the-art in power magnetics modeling; (2) set up examples for fostering an open-source and transparent research community; (3) developed useful guidelines and practical rules for conducting data-driven research in power electronics; and (4) provided a fair performance benchmark leading to insights on the most promising future research directions. The competition yielded a collection of publicly disclosed software algorithms and tools designed to capture the distinct loss characteristics of power magnetic materials, which are mostly open-sourced. We have attempted to bridge power electronics domain knowledge with state-of-the-art advancements in artificial intelligence, machine learning, pattern recognition, and signal processing. The MagNet Challenge has greatly improved the accuracy and reduced the size of data-driven power magnetic material models. The models and tools created for various materials were meticulously documented and shared within the broader power electronics community.